llvm dev - Sep 2014 - [LLVMdev] Optimization of sqrt() with invalid argument

> On Sep 25, 2014, at 6:41 PM, Tim Northover <t.p.northover at
gmail.com> wrote:
> 
>> This is important in practice because of the xalanc benchmark in SPEC
>> CPU2006.
> 
> No, it would be important in practice for really used legacy (i.e.
> unfixable) code relying on this undefined behaviour. It's important
> for gaming the system if it occurs in SPEC.
We should fix it because it costs us nothing to do and conforms to IEEE-754, and
there’s no good reason to maintain the current behavior.

– Steve

Bill raised another question in the bug report that I'm curious about too:
Is anyone or any bot running/posting SPEC CPU results for clang?

I know it's part of test-suite/external, but this constant fold code has
been around 5+ years. Was the bug lying dormant all this time, only visible
on PPC, or something else?
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140926/5c2bbf9d/attachment.html>

> We should fix it because it costs us nothing to do and conforms to
IEEE-754, and there’s no good reason to maintain the current behavior.
Probably, but there are theoretical issues: 0 is easier for other
optimisations to deal with than NaN (especially after constant
propagation). It's also creating NaNs in situations where they've been
explicitly disabled (the @llvm.sqrt intrinsic only really gets created
in finite math mode). In my perfect world we'd optimise it to a
segfault.

I doubt anything bad will actually happen though, so I'm not actually
bothered one way or the other; just vaguely annoyed that compilers get
changed to accommodate broken benchmarks (well, more accurately, to
allow us to sneak a -ffast-math compile into our numbers when the
benchmark wasn't designed for it).

Cheers.

Tim.

> I know it's part of test-suite/external, but this constant fold code
has
> been around 5+ years. Was the bug lying dormant all this time, only visible
> on PPC, or something else?
My guess would be that people don't tend to run with -ffast-math (it's
got a reputation for breaking code, even ignoring this particular
issue). Without that, from what I can see the issue won't arise.

But that doesn't account for complete silence, it must happen occasionally.

Cheers.

Tim.